Security and privacy

There are two ways to use Complision, and each has a different privacy model. Here it is stated plainly, not buried in small print.

Two models, depending on what you use

Your file never leaves your browser. The document itself is never uploaded to any server, ours or anyone else’s.
To classify each question against our control catalog, the text of the question is processed — only the text. Never the document, the file name, your company’s name, or any answer you’ve already filled in. Security questionnaire questions are generic: hundreds of companies get the same one, so the text alone identifies no one.
That text is first checked against questions already classified before (so the work isn’t repeated) and, only if needed, sent to Claude’s API (Anthropic) exclusively for that classification — never to generate content or for any other use.
Close the tab and nothing of yours remains. No account, no session, no database holding your questionnaire or which company sent it.
We only store anything if you explicitly ask for it: the aggregated result (number of questions, gaps, detected frameworks), never the file’s content, when you download the report or request the service.

Once you hire the service, you upload your evidence (policies, screenshots, reports) through a temporary link that expires in 7 days.
We process each piece of evidence — classify it, extract its date, calculate when it expires — and generate your answer and evidence index.
The original file is deleted after 30 days. What remains afterward is: metadata (normalized name, date, type), a SHA-256 verification hash, the already-generated answer, and the control’s status. Not the document itself.
Why we do it this way: permanently storing the security-weakness map of dozens of companies would make us a target, and the damage from a breach would be yours, not ours. We’d rather not accumulate that risk.

We store

Metadata for each piece of evidence: reference, type, normalized name, document date, calculated expiry date
A SHA-256 hash of each file, so we can verify its integrity without needing to keep it
The generated answer for each question and its status (draft, reviewed, approved, gap)
An audit log of who accessed what and when

We don’t store

The original file of any evidence, past 30 days
The questionnaire you upload to the free triage — never, not even during analysis
Passwords: access to the client area is via magic link, so there’s no password to leak

Every organization is isolated at the database level (Row Level Security), not just by application logic — a code bug can’t leak one company’s data into another’s.
If one of our operators needs to see a specific document to resolve a question, they access it via a single-use signed URL that expires in 15 minutes, and that access is logged: who, what, when, from which IP. No exceptions.
Data is hosted on European infrastructure (Frankfurt or Ireland), never outside the European Economic Area.
Internal admin access requires two-factor authentication.

We use Plausible, a cookieless analytics tool: it doesn’t identify individual visitors or track you across sites.
We only log aggregated funnel events — which page gets visited, where in the analysis flow people drop off, whether the report gets downloaded — never your questionnaire’s text, the file name, your company’s name, or your email.
Hosted in the European Union. That’s why there’s no consent banner when you arrive: there’s no personal data to consent to.